Anyone who thinks vision has already largely been solved in AI is kidding themselves. Object recognition ≠ scene comprehension.https://twitter.com/skeptoptimist/status/1031808308154261504 …
-
right, I took supervised learning to be a given - it's obviously not easy to learn unsupervised. In any case, this whole notion of 'comparable to human learning' is besides the point to me; nothing in AI is truly comparable, Atari included. The whole criticism is poorly motivated
no; humans represent an important benchmark. DoTA work is still interesting, but framing it as a competition with humans is misleading to the extent that they are playing different games.
-
-
'Comparable to human learning' and 'comparable to human performance given same inputs' is not the same thing. As I said, if all you want is inputs to be the same, then supervised learning + engineering can learn the pixels -> mapping, almost certainly.
-
As others noted learning pixels->mapping given no sim supervision is MUCH harder, but that bit is unrelated to objection that "framing it as aa competition with humans is misleading to the extent that they are playing different games."
- 1 more reply
New conversation -
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.